multipathd/main.c

   1 /*
   2  * Copyright (c) 2004, 2005 Christophe Varoqui
   3  * Copyright (c) 2005 Kiyoshi Ueda, NEC
   4  * Copyright (c) 2005 Benjamin Marzinski, Redhat
   5  * Copyright (c) 2005 Edward Goggin, EMC
   6  */
   7 #include <unistd.h>
   8 #include <sys/stat.h>
   9 #include <libdevmapper.h>
  10 #include <wait.h>
  11 #include <sys/mman.h>
  12 #include <sys/types.h>
  13 #include <fcntl.h>
  14 #include <errno.h>
  15 #include <sys/time.h>
  16 #include <sys/resource.h>
  17 #include <limits.h>
  18 #include <linux/oom.h>
  19 #include <libudev.h>
  20 #ifdef USE_SYSTEMD
  21 #include <systemd/sd-daemon.h>
  22 #endif
  23 #include <semaphore.h>
  24 #include <mpath_persist.h>
  25
  26 /*
  27  * libcheckers
  28  */
  29 #include <checkers.h>
  30
  31 /*
  32  * libmultipath
  33  */
  34 #include <parser.h>
  35 #include <vector.h>
  36 #include <memory.h>
  37 #include <config.h>
  38 #include <util.h>
  39 #include <hwtable.h>
  40 #include <defaults.h>
  41 #include <structs.h>
  42 #include <blacklist.h>
  43 #include <structs_vec.h>
  44 #include <dmparser.h>
  45 #include <devmapper.h>
  46 #include <sysfs.h>
  47 #include <dict.h>
  48 #include <discovery.h>
  49 #include <debug.h>
  50 #include <propsel.h>
  51 #include <uevent.h>
  52 #include <switchgroup.h>
  53 #include <print.h>
  54 #include <configure.h>
  55 #include <prio.h>
  56 #include <pgpolicies.h>
  57 #include <uevent.h>
  58 #include <log.h>
  59
  60 #include "main.h"
  61 #include "pidfile.h"
  62 #include "uxlsnr.h"
  63 #include "uxclnt.h"
  64 #include "cli.h"
  65 #include "cli_handlers.h"
  66 #include "lock.h"
  67 #include "waiter.h"
  68 #include "wwids.h"
  69
  70 #define FILE_NAME_SIZE 256
  71 #define CMDSIZE 160
  72
  73 #define LOG_MSG(a, b) \
  74 do { \
  75         if (pp->offline) \
  76                 condlog(a, "%s: %s - path offline", pp->mpp->alias, pp->dev); \
  77         else if (strlen(b)) \
  78                 condlog(a, "%s: %s - %s", pp->mpp->alias, pp->dev, b); \
  79 } while(0)
  80
  81 struct mpath_event_param
  82 {
  83         char * devname;
  84         struct multipath *mpp;
  85 };
  86
  87 unsigned int mpath_mx_alloc_len;
  88
  89 int logsink;
  90 enum daemon_status running_state;
  91 pid_t daemon_pid;
  92
  93 static sem_t exit_sem;
  94 /*
  95  * global copy of vecs for use in sig handlers
  96  */
  97 struct vectors * gvecs;
  98
  99 struct udev * udev;
 100
 101 static int
 102 need_switch_pathgroup (struct multipath * mpp, int refresh)
 103 {
 104         struct pathgroup * pgp;
 105         struct path * pp;
 106         unsigned int i, j;
 107
 108         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
 109                 return 0;
 110
 111         /*
 112          * Refresh path priority values
 113          */
 114         if (refresh)
 115                 vector_foreach_slot (mpp->pg, pgp, i)
 116                         vector_foreach_slot (pgp->paths, pp, j)
 117                                 pathinfo(pp, conf->hwtable, DI_PRIO);
 118
 119         mpp->bestpg = select_path_group(mpp);
 120
 121         if (mpp->bestpg != mpp->nextpg)
 122                 return 1;
 123
 124         return 0;
 125 }
 126
 127 static void
 128 switch_pathgroup (struct multipath * mpp)
 129 {
 130         mpp->stat_switchgroup++;
 131         dm_switchgroup(mpp->alias, mpp->bestpg);
 132         condlog(2, "%s: switch to path group #%i",
 133                  mpp->alias, mpp->bestpg);
 134 }
 135
 136 static int
 137 coalesce_maps(struct vectors *vecs, vector nmpv)
 138 {
 139         struct multipath * ompp;
 140         vector ompv = vecs->mpvec;
 141         unsigned int i;
 142
 143         vector_foreach_slot (ompv, ompp, i) {
 144                 condlog(3, "%s: coalesce map", ompp->alias);
 145                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 146                         /*
 147                          * remove all current maps not allowed by the
 148                          * current configuration
 149                          */
 150                         if (dm_flush_map(ompp->alias)) {
 151                                 condlog(0, "%s: unable to flush devmap",
 152                                         ompp->alias);
 153                                 /*
 154                                  * may be just because the device is open
 155                                  */
 156                                 if (setup_multipath(vecs, ompp) != 0) {
 157                                         i--;
 158                                         continue;
 159                                 }
 160                                 if (!vector_alloc_slot(nmpv))
 161                                         return 1;
 162
 163                                 vector_set_slot(nmpv, ompp);
 164
 165                                 vector_del_slot(ompv, i);
 166                                 i--;
 167                         }
 168                         else {
 169                                 dm_lib_release();
 170                                 condlog(2, "%s devmap removed", ompp->alias);
 171                         }
 172                 } else if (conf->reassign_maps) {
 173                         condlog(3, "%s: Reassign existing device-mapper"
 174                                 " devices", ompp->alias);
 175                         dm_reassign(ompp->alias);
 176                 }
 177         }
 178         return 0;
 179 }
 180
 181 void
 182 sync_map_state(struct multipath *mpp)
 183 {
 184         struct pathgroup *pgp;
 185         struct path *pp;
 186         unsigned int i, j;
 187
 188         if (!mpp->pg)
 189                 return;
 190
 191         vector_foreach_slot (mpp->pg, pgp, i){
 192                 vector_foreach_slot (pgp->paths, pp, j){
 193                         if (pp->state == PATH_UNCHECKED ||
 194                             pp->state == PATH_WILD)
 195                                 continue;
 196                         if ((pp->dmstate == PSTATE_FAILED ||
 197                              pp->dmstate == PSTATE_UNDEF) &&
 198                             (pp->state == PATH_UP || pp->state == PATH_GHOST))
 199                                 dm_reinstate_path(mpp->alias, pp->dev_t);
 200                         else if ((pp->dmstate == PSTATE_ACTIVE ||
 201                                   pp->dmstate == PSTATE_UNDEF) &&
 202                                  (pp->state == PATH_DOWN ||
 203                                   pp->state == PATH_SHAKY))
 204                                 dm_fail_path(mpp->alias, pp->dev_t);
 205                 }
 206         }
 207 }
 208
 209 static void
 210 sync_maps_state(vector mpvec)
 211 {
 212         unsigned int i;
 213         struct multipath *mpp;
 214
 215         vector_foreach_slot (mpvec, mpp, i)
 216                 sync_map_state(mpp);
 217 }
 218
 219 static int
 220 flush_map(struct multipath * mpp, struct vectors * vecs)
 221 {
 222         /*
 223          * clear references to this map before flushing so we can ignore
 224          * the spurious uevent we may generate with the dm_flush_map call below
 225          */
 226         if (dm_flush_map(mpp->alias)) {
 227                 /*
 228                  * May not really be an error -- if the map was already flushed
 229                  * from the device mapper by dmsetup(8) for instance.
 230                  */
 231                 condlog(0, "%s: can't flush", mpp->alias);
 232                 return 1;
 233         }
 234         else {
 235                 dm_lib_release();
 236                 condlog(2, "%s: map flushed", mpp->alias);
 237         }
 238
 239         orphan_paths(vecs->pathvec, mpp);
 240         remove_map_and_stop_waiter(mpp, vecs, 1);
 241
 242         return 0;
 243 }
 244
 245 static int
 246 uev_add_map (struct uevent * uev, struct vectors * vecs)
 247 {
 248         char *alias;
 249         int major = -1, minor = -1, rc;
 250
 251         condlog(3, "%s: add map (uevent)", uev->kernel);
 252         alias = uevent_get_dm_name(uev);
 253         if (!alias) {
 254                 condlog(3, "%s: No DM_NAME in uevent", uev->kernel);
 255                 major = uevent_get_major(uev);
 256                 minor = uevent_get_minor(uev);
 257                 alias = dm_mapname(major, minor);
 258                 if (!alias) {
 259                         condlog(2, "%s: mapname not found for %d:%d",
 260                                 uev->kernel, major, minor);
 261                         return 1;
 262                 }
 263         }
 264         rc = ev_add_map(uev->kernel, alias, vecs);
 265         FREE(alias);
 266         return rc;
 267 }
 268
 269 int
 270 ev_add_map (char * dev, char * alias, struct vectors * vecs)
 271 {
 272         char * refwwid;
 273         struct multipath * mpp;
 274         int map_present;
 275         int r = 1;
 276
 277         map_present = dm_map_present(alias);
 278
 279         if (map_present && dm_type(alias, TGT_MPATH) <= 0) {
 280                 condlog(4, "%s: not a multipath map", alias);
 281                 return 0;
 282         }
 283
 284         mpp = find_mp_by_alias(vecs->mpvec, alias);
 285
 286         if (mpp) {
 287                 /*
 288                  * Not really an error -- we generate our own uevent
 289                  * if we create a multipath mapped device as a result
 290                  * of uev_add_path
 291                  */
 292                 if (conf->reassign_maps) {
 293                         condlog(3, "%s: Reassign existing device-mapper devices",
 294                                 alias);
 295                         dm_reassign(alias);
 296                 }
 297                 return 0;
 298         }
 299         condlog(2, "%s: adding map", alias);
 300
 301         /*
 302          * now we can register the map
 303          */
 304         if (map_present && (mpp = add_map_without_path(vecs, alias))) {
 305                 sync_map_state(mpp);
 306                 condlog(2, "%s: devmap %s registered", alias, dev);
 307                 return 0;
 308         }
 309         r = get_refwwid(dev, DEV_DEVMAP, vecs->pathvec, &refwwid);
 310
 311         if (refwwid) {
 312                 r = coalesce_paths(vecs, NULL, refwwid, 0);
 313                 dm_lib_release();
 314         }
 315
 316         if (!r)
 317                 condlog(2, "%s: devmap %s added", alias, dev);
 318         else if (r == 2)
 319                 condlog(2, "%s: uev_add_map %s blacklisted", alias, dev);
 320         else
 321                 condlog(0, "%s: uev_add_map %s failed", alias, dev);
 322
 323         FREE(refwwid);
 324         return r;
 325 }
 326
 327 static int
 328 uev_remove_map (struct uevent * uev, struct vectors * vecs)
 329 {
 330         char *alias;
 331         int minor;
 332         struct multipath *mpp;
 333
 334         condlog(2, "%s: remove map (uevent)", uev->kernel);
 335         alias = uevent_get_dm_name(uev);
 336         if (!alias) {
 337                 condlog(3, "%s: No DM_NAME in uevent, ignoring", uev->kernel);
 338                 return 0;
 339         }
 340         minor = uevent_get_minor(uev);
 341         mpp = find_mp_by_minor(vecs->mpvec, minor);
 342
 343         if (!mpp) {
 344                 condlog(2, "%s: devmap not registered, can't remove",
 345                         uev->kernel);
 346                 goto out;
 347         }
 348         if (strcmp(mpp->alias, alias)) {
 349                 condlog(2, "%s: minor number mismatch (map %d, event %d)",
 350                         mpp->alias, mpp->dmi->minor, minor);
 351                 goto out;
 352         }
 353
 354         orphan_paths(vecs->pathvec, mpp);
 355         remove_map_and_stop_waiter(mpp, vecs, 1);
 356 out:
 357         FREE(alias);
 358         return 0;
 359 }
 360
 361 int
 362 ev_remove_map (char * devname, char * alias, int minor, struct vectors * vecs)
 363 {
 364         struct multipath * mpp;
 365
 366         mpp = find_mp_by_minor(vecs->mpvec, minor);
 367
 368         if (!mpp) {
 369                 condlog(2, "%s: devmap not registered, can't remove",
 370                         devname);
 371                 return 0;
 372         }
 373         if (strcmp(mpp->alias, alias)) {
 374                 condlog(2, "%s: minor number mismatch (map %d, event %d)",
 375                         mpp->alias, mpp->dmi->minor, minor);
 376                 return 0;
 377         }
 378         return flush_map(mpp, vecs);
 379 }
 380
 381 static int
 382 uev_add_path (struct uevent *uev, struct vectors * vecs)
 383 {
 384         struct path *pp;
 385         int ret, i;
 386
 387         condlog(2, "%s: add path (uevent)", uev->kernel);
 388         if (strstr(uev->kernel, "..") != NULL) {
 389                 /*
 390                  * Don't allow relative device names in the pathvec
 391                  */
 392                 condlog(0, "%s: path name is invalid", uev->kernel);
 393                 return 1;
 394         }
 395
 396         pp = find_path_by_dev(vecs->pathvec, uev->kernel);
 397         if (pp) {
 398                 condlog(0, "%s: spurious uevent, path already in pathvec",
 399                         uev->kernel);
 400                 if (pp->mpp)
 401                         return 0;
 402                 if (!strlen(pp->wwid)) {
 403                         udev_device_unref(pp->udev);
 404                         pp->udev = udev_device_ref(uev->udev);
 405                         ret = pathinfo(pp, conf->hwtable,
 406                                        DI_ALL | DI_BLACKLIST);
 407                         if (ret == 2) {
 408                                 i = find_slot(vecs->pathvec, (void *)pp);
 409                                 if (i != -1)
 410                                         vector_del_slot(vecs->pathvec, i);
 411                                 free_path(pp);
 412                                 return 0;
 413                         } else if (ret == 1) {
 414                                 condlog(0, "%s: failed to reinitialize path",
 415                                         uev->kernel);
 416                                 return 1;
 417                         }
 418                 }
 419         } else {
 420                 /*
 421                  * get path vital state
 422                  */
 423                 ret = store_pathinfo(vecs->pathvec, conf->hwtable,
 424                                      uev->udev, DI_ALL, &pp);
 425                 if (!pp) {
 426                         if (ret == 2)
 427                                 return 0;
 428                         condlog(0, "%s: failed to store path info",
 429                                 uev->kernel);
 430                         return 1;
 431                 }
 432                 pp->checkint = conf->checkint;
 433         }
 434
 435         return ev_add_path(pp, vecs);
 436 }
 437
 438 /*
 439  * returns:
 440  * 0: added
 441  * 1: error
 442  */
 443 int
 444 ev_add_path (struct path * pp, struct vectors * vecs)
 445 {
 446         struct multipath * mpp;
 447         char empty_buff[WWID_SIZE] = {0};
 448         char params[PARAMS_SIZE] = {0};
 449         int retries = 3;
 450         int start_waiter = 0;
 451
 452         /*
 453          * need path UID to go any further
 454          */
 455         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 456                 condlog(0, "%s: failed to get path uid", pp->dev);
 457                 goto fail; /* leave path added to pathvec */
 458         }
 459         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 460 rescan:
 461         if (mpp) {
 462                 if ((!pp->size) || (mpp->size != pp->size)) {
 463                         if (!pp->size)
 464                                 condlog(0, "%s: failed to add new path %s, "
 465                                         "device size is 0",
 466                                         mpp->alias, pp->dev);
 467                         else
 468                                 condlog(0, "%s: failed to add new path %s, "
 469                                         "device size mismatch",
 470                                         mpp->alias, pp->dev);
 471                         int i = find_slot(vecs->pathvec, (void *)pp);
 472                         if (i != -1)
 473                                 vector_del_slot(vecs->pathvec, i);
 474                         free_path(pp);
 475                         return 1;
 476                 }
 477
 478                 condlog(4,"%s: adopting all paths for path %s",
 479                         mpp->alias, pp->dev);
 480                 if (adopt_paths(vecs->pathvec, mpp, 1))
 481                         goto fail; /* leave path added to pathvec */
 482
 483                 verify_paths(mpp, vecs, NULL);
 484                 mpp->flush_on_last_del = FLUSH_UNDEF;
 485                 mpp->action = ACT_RELOAD;
 486         }
 487         else {
 488                 if (!pp->size) {
 489                         condlog(0, "%s: failed to create new map,"
 490                                 " device size is 0 ", pp->dev);
 491                         int i = find_slot(vecs->pathvec, (void *)pp);
 492                         if (i != -1)
 493                                 vector_del_slot(vecs->pathvec, i);
 494                         free_path(pp);
 495                         return 1;
 496                 }
 497
 498                 condlog(4,"%s: creating new map", pp->dev);
 499                 if ((mpp = add_map_with_path(vecs, pp, 1))) {
 500                         mpp->action = ACT_CREATE;
 501                         /*
 502                          * We don't depend on ACT_CREATE, as domap will
 503                          * set it to ACT_NOTHING when complete.
 504                          */
 505                         start_waiter = 1;
 506                 }
 507                 else
 508                         goto fail; /* leave path added to pathvec */
 509         }
 510
 511         /* persistent reseravtion check*/
 512         mpath_pr_event_handle(pp);
 513
 514         /*
 515          * push the map to the device-mapper
 516          */
 517         if (setup_map(mpp, params, PARAMS_SIZE)) {
 518                 condlog(0, "%s: failed to setup map for addition of new "
 519                         "path %s", mpp->alias, pp->dev);
 520                 goto fail_map;
 521         }
 522         /*
 523          * reload the map for the multipath mapped device
 524          */
 525         if (domap(mpp, params) <= 0) {
 526                 condlog(0, "%s: failed in domap for addition of new "
 527                         "path %s", mpp->alias, pp->dev);
 528                 /*
 529                  * deal with asynchronous uevents :((
 530                  */
 531                 if (mpp->action == ACT_RELOAD && retries-- > 0) {
 532                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 533                         sleep(1);
 534                         update_mpp_paths(mpp, vecs->pathvec);
 535                         goto rescan;
 536                 }
 537                 else if (mpp->action == ACT_RELOAD)
 538                         condlog(0, "%s: giving up reload", mpp->alias);
 539                 else
 540                         goto fail_map;
 541         }
 542         dm_lib_release();
 543
 544         /*
 545          * update our state from kernel regardless of create or reload
 546          */
 547         if (setup_multipath(vecs, mpp))
 548                 goto fail; /* if setup_multipath fails, it removes the map */
 549
 550         sync_map_state(mpp);
 551
 552         if ((mpp->action == ACT_CREATE ||
 553              (mpp->action == ACT_NOTHING && start_waiter && !mpp->waiter)) &&
 554             start_waiter_thread(mpp, vecs))
 555                         goto fail_map;
 556
 557         if (retries >= 0) {
 558                 condlog(2, "%s [%s]: path added to devmap %s",
 559                         pp->dev, pp->dev_t, mpp->alias);
 560                 return 0;
 561         }
 562         else
 563                 return 1;
 564
 565 fail_map:
 566         remove_map(mpp, vecs, 1);
 567 fail:
 568         orphan_path(pp, "failed to add path");
 569         return 1;
 570 }
 571
 572 static int
 573 uev_remove_path (struct uevent *uev, struct vectors * vecs)
 574 {
 575         struct path *pp;
 576
 577         condlog(2, "%s: remove path (uevent)", uev->kernel);
 578         pp = find_path_by_dev(vecs->pathvec, uev->kernel);
 579
 580         if (!pp) {
 581                 /* Not an error; path might have been purged earlier */
 582                 condlog(0, "%s: path already removed", uev->kernel);
 583                 return 0;
 584         }
 585
 586         return ev_remove_path(pp, vecs);
 587 }
 588
 589 int
 590 ev_remove_path (struct path *pp, struct vectors * vecs)
 591 {
 592         struct multipath * mpp;
 593         int i, retval = 0;
 594         char params[PARAMS_SIZE] = {0};
 595
 596         /*
 597          * avoid referring to the map of an orphaned path
 598          */
 599         if ((mpp = pp->mpp)) {
 600                 /*
 601                  * transform the mp->pg vector of vectors of paths
 602                  * into a mp->params string to feed the device-mapper
 603                  */
 604                 if (update_mpp_paths(mpp, vecs->pathvec)) {
 605                         condlog(0, "%s: failed to update paths",
 606                                 mpp->alias);
 607                         goto fail;
 608                 }
 609                 if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 610                         vector_del_slot(mpp->paths, i);
 611
 612                 /*
 613                  * remove the map IFF removing the last path
 614                  */
 615                 if (VECTOR_SIZE(mpp->paths) == 0) {
 616                         char alias[WWID_SIZE];
 617
 618                         /*
 619                          * flush_map will fail if the device is open
 620                          */
 621                         strncpy(alias, mpp->alias, WWID_SIZE);
 622                         if (mpp->flush_on_last_del == FLUSH_ENABLED) {
 623                                 condlog(2, "%s Last path deleted, disabling queueing", mpp->alias);
 624                                 mpp->retry_tick = 0;
 625                                 mpp->no_path_retry = NO_PATH_RETRY_FAIL;
 626                                 mpp->flush_on_last_del = FLUSH_IN_PROGRESS;
 627                                 dm_queue_if_no_path(mpp->alias, 0);
 628                         }
 629                         if (!flush_map(mpp, vecs)) {
 630                                 condlog(2, "%s: removed map after"
 631                                         " removing all paths",
 632                                         alias);
 633                                 retval = 0;
 634                                 goto out;
 635                         }
 636                         /*
 637                          * Not an error, continue
 638                          */
 639                 }
 640
 641                 if (setup_map(mpp, params, PARAMS_SIZE)) {
 642                         condlog(0, "%s: failed to setup map for"
 643                                 " removal of path %s", mpp->alias, pp->dev);
 644                         goto fail;
 645                 }
 646                 /*
 647                  * reload the map
 648                  */
 649                 mpp->action = ACT_RELOAD;
 650                 if (domap(mpp, params) <= 0) {
 651                         condlog(0, "%s: failed in domap for "
 652                                 "removal of path %s",
 653                                 mpp->alias, pp->dev);
 654                         retval = 1;
 655                 } else {
 656                         /*
 657                          * update our state from kernel
 658                          */
 659                         if (setup_multipath(vecs, mpp)) {
 660                                 goto fail;
 661                         }
 662                         sync_map_state(mpp);
 663
 664                         condlog(2, "%s [%s]: path removed from map %s",
 665                                 pp->dev, pp->dev_t, mpp->alias);
 666                 }
 667         }
 668
 669 out:
 670         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 671                 vector_del_slot(vecs->pathvec, i);
 672
 673         free_path(pp);
 674
 675         return retval;
 676
 677 fail:
 678         remove_map_and_stop_waiter(mpp, vecs, 1);
 679         return 1;
 680 }
 681
 682 static int
 683 uev_update_path (struct uevent *uev, struct vectors * vecs)
 684 {
 685         int ro, retval = 0;
 686
 687         ro = uevent_get_disk_ro(uev);
 688
 689         if (ro >= 0) {
 690                 struct path * pp;
 691
 692                 condlog(2, "%s: update path write_protect to '%d' (uevent)",
 693                         uev->kernel, ro);
 694                 pp = find_path_by_dev(vecs->pathvec, uev->kernel);
 695                 if (!pp) {
 696                         condlog(0, "%s: spurious uevent, path not found",
 697                                 uev->kernel);
 698                         return 1;
 699                 }
 700                 if (pp->mpp) {
 701                         retval = reload_map(vecs, pp->mpp, 0);
 702
 703                         condlog(2, "%s: map %s reloaded (retval %d)",
 704                                 uev->kernel, pp->mpp->alias, retval);
 705                 }
 706
 707         }
 708
 709         return retval;
 710 }
 711
 712 static int
 713 map_discovery (struct vectors * vecs)
 714 {
 715         struct multipath * mpp;
 716         unsigned int i;
 717
 718         if (dm_get_maps(vecs->mpvec))
 719                 return 1;
 720
 721         vector_foreach_slot (vecs->mpvec, mpp, i)
 722                 if (setup_multipath(vecs, mpp))
 723                         return 1;
 724
 725         return 0;
 726 }
 727
 728 int
 729 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 730 {
 731         struct vectors * vecs;
 732         int r;
 733
 734         *reply = NULL;
 735         *len = 0;
 736         vecs = (struct vectors *)trigger_data;
 737
 738         pthread_cleanup_push(cleanup_lock, &vecs->lock);
 739         lock(vecs->lock);
 740         pthread_testcancel();
 741
 742         r = parse_cmd(str, reply, len, vecs);
 743
 744         if (r > 0) {
 745                 *reply = STRDUP("fail\n");
 746                 *len = strlen(*reply) + 1;
 747                 r = 1;
 748         }
 749         else if (!r && *len == 0) {
 750                 *reply = STRDUP("ok\n");
 751                 *len = strlen(*reply) + 1;
 752                 r = 0;
 753         }
 754         /* else if (r < 0) leave *reply alone */
 755
 756         lock_cleanup_pop(vecs->lock);
 757         return r;
 758 }
 759
 760 static int
 761 uev_discard(char * devpath)
 762 {
 763         char *tmp;
 764         char a[11], b[11];
 765
 766         /*
 767          * keep only block devices, discard partitions
 768          */
 769         tmp = strstr(devpath, "/block/");
 770         if (tmp == NULL){
 771                 condlog(4, "no /block/ in '%s'", devpath);
 772                 return 1;
 773         }
 774         if (sscanf(tmp, "/block/%10s", a) != 1 ||
 775             sscanf(tmp, "/block/%10[^/]/%10s", a, b) == 2) {
 776                 condlog(4, "discard event on %s", devpath);
 777                 return 1;
 778         }
 779         return 0;
 780 }
 781
 782 int
 783 uev_trigger (struct uevent * uev, void * trigger_data)
 784 {
 785         int r = 0;
 786         struct vectors * vecs;
 787
 788         vecs = (struct vectors *)trigger_data;
 789
 790         if (uev_discard(uev->devpath))
 791                 return 0;
 792
 793         pthread_cleanup_push(cleanup_lock, &vecs->lock);
 794         lock(vecs->lock);
 795         pthread_testcancel();
 796
 797         /*
 798          * device map event
 799          * Add events are ignored here as the tables
 800          * are not fully initialised then.
 801          */
 802         if (!strncmp(uev->kernel, "dm-", 3)) {
 803                 if (!strncmp(uev->action, "change", 6)) {
 804                         r = uev_add_map(uev, vecs);
 805                         goto out;
 806                 }
 807                 if (!strncmp(uev->action, "remove", 6)) {
 808                         r = uev_remove_map(uev, vecs);
 809                         goto out;
 810                 }
 811                 goto out;
 812         }
 813
 814         /*
 815          * path add/remove event
 816          */
 817         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 818                            uev->kernel) > 0)
 819                 goto out;
 820
 821         if (!strncmp(uev->action, "add", 3)) {
 822                 r = uev_add_path(uev, vecs);
 823                 goto out;
 824         }
 825         if (!strncmp(uev->action, "remove", 6)) {
 826                 r = uev_remove_path(uev, vecs);
 827                 goto out;
 828         }
 829         if (!strncmp(uev->action, "change", 6)) {
 830                 r = uev_update_path(uev, vecs);
 831                 goto out;
 832         }
 833
 834 out:
 835         lock_cleanup_pop(vecs->lock);
 836         return r;
 837 }
 838
 839 static void *
 840 ueventloop (void * ap)
 841 {
 842         if (uevent_listen(udev))
 843                 condlog(0, "error starting uevent listener");
 844
 845         return NULL;
 846 }
 847
 848 static void *
 849 uevqloop (void * ap)
 850 {
 851         if (uevent_dispatch(&uev_trigger, ap))
 852                 condlog(0, "error starting uevent dispatcher");
 853
 854         return NULL;
 855 }
 856 static void *
 857 uxlsnrloop (void * ap)
 858 {
 859         if (cli_init())
 860                 return NULL;
 861
 862         set_handler_callback(LIST+PATHS, cli_list_paths);
 863         set_handler_callback(LIST+PATHS+FMT, cli_list_paths_fmt);
 864         set_handler_callback(LIST+MAPS, cli_list_maps);
 865         set_handler_callback(LIST+STATUS, cli_list_status);
 866         set_handler_callback(LIST+DAEMON, cli_list_daemon);
 867         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 868         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 869         set_handler_callback(LIST+MAPS+FMT, cli_list_maps_fmt);
 870         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 871         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 872         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 873         set_handler_callback(LIST+CONFIG, cli_list_config);
 874         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 875         set_handler_callback(LIST+DEVICES, cli_list_devices);
 876         set_handler_callback(LIST+WILDCARDS, cli_list_wildcards);
 877         set_handler_callback(ADD+PATH, cli_add_path);
 878         set_handler_callback(DEL+PATH, cli_del_path);
 879         set_handler_callback(ADD+MAP, cli_add_map);
 880         set_handler_callback(DEL+MAP, cli_del_map);
 881         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 882         set_handler_callback(RECONFIGURE, cli_reconfigure);
 883         set_handler_callback(SUSPEND+MAP, cli_suspend);
 884         set_handler_callback(RESUME+MAP, cli_resume);
 885         set_handler_callback(RESIZE+MAP, cli_resize);
 886         set_handler_callback(RELOAD+MAP, cli_reload);
 887         set_handler_callback(RESET+MAP, cli_reassign);
 888         set_handler_callback(REINSTATE+PATH, cli_reinstate);
 889         set_handler_callback(FAIL+PATH, cli_fail);
 890         set_handler_callback(DISABLEQ+MAP, cli_disable_queueing);
 891         set_handler_callback(RESTOREQ+MAP, cli_restore_queueing);
 892         set_handler_callback(DISABLEQ+MAPS, cli_disable_all_queueing);
 893         set_handler_callback(RESTOREQ+MAPS, cli_restore_all_queueing);
 894         set_handler_callback(QUIT, cli_quit);
 895         set_handler_callback(SHUTDOWN, cli_shutdown);
 896         set_handler_callback(GETPRSTATUS+MAP, cli_getprstatus);
 897         set_handler_callback(SETPRSTATUS+MAP, cli_setprstatus);
 898         set_handler_callback(UNSETPRSTATUS+MAP, cli_unsetprstatus);
 899         set_handler_callback(FORCEQ+DAEMON, cli_force_no_daemon_q);
 900         set_handler_callback(RESTOREQ+DAEMON, cli_restore_no_daemon_q);
 901
 902         umask(077);
 903         uxsock_listen(&uxsock_trigger, ap);
 904
 905         return NULL;
 906 }
 907
 908 void
 909 exit_daemon (void)
 910 {
 911         sem_post(&exit_sem);
 912 }
 913
 914 const char *
 915 daemon_status(void)
 916 {
 917         switch (running_state) {
 918         case DAEMON_INIT:
 919                 return "init";
 920         case DAEMON_START:
 921                 return "startup";
 922         case DAEMON_CONFIGURE:
 923                 return "configure";
 924         case DAEMON_RUNNING:
 925                 return "running";
 926         case DAEMON_SHUTDOWN:
 927                 return "shutdown";
 928         }
 929         return NULL;
 930 }
 931
 932 static void
 933 fail_path (struct path * pp, int del_active)
 934 {
 935         if (!pp->mpp)
 936                 return;
 937
 938         condlog(2, "checker failed path %s in map %s",
 939                  pp->dev_t, pp->mpp->alias);
 940
 941         dm_fail_path(pp->mpp->alias, pp->dev_t);
 942         if (del_active)
 943                 update_queue_mode_del_path(pp->mpp);
 944 }
 945
 946 /*
 947  * caller must have locked the path list before calling that function
 948  */
 949 static void
 950 reinstate_path (struct path * pp, int add_active)
 951 {
 952         if (!pp->mpp)
 953                 return;
 954
 955         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 956                 condlog(0, "%s: reinstate failed", pp->dev_t);
 957         else {
 958                 condlog(2, "%s: reinstated", pp->dev_t);
 959                 if (add_active)
 960                         update_queue_mode_add_path(pp->mpp);
 961         }
 962 }
 963
 964 static void
 965 enable_group(struct path * pp)
 966 {
 967         struct pathgroup * pgp;
 968
 969         /*
 970          * if path is added through uev_add_path, pgindex can be unset.
 971          * next update_strings() will set it, upon map reload event.
 972          *
 973          * we can safely return here, because upon map reload, all
 974          * PG will be enabled.
 975          */
 976         if (!pp->mpp->pg || !pp->pgindex)
 977                 return;
 978
 979         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 980
 981         if (pgp->status == PGSTATE_DISABLED) {
 982                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 983                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 984         }
 985 }
 986
 987 static void
 988 mpvec_garbage_collector (struct vectors * vecs)
 989 {
 990         struct multipath * mpp;
 991         unsigned int i;
 992
 993         if (!vecs->mpvec)
 994                 return;
 995
 996         vector_foreach_slot (vecs->mpvec, mpp, i) {
 997                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
 998                         condlog(2, "%s: remove dead map", mpp->alias);
 999                         remove_map_and_stop_waiter(mpp, vecs, 1);
1000                         i--;
1001                 }
1002         }
1003 }
1004
1005 /* This is called after a path has started working again. It the multipath
1006  * device for this path uses the followover failback type, and this is the
1007  * best pathgroup, and this is the first path in the pathgroup to come back
1008  * up, then switch to this pathgroup */
1009 static int
1010 followover_should_failback(struct path * pp)
1011 {
1012         struct pathgroup * pgp;
1013         struct path *pp1;
1014         int i;
1015
1016         if (pp->mpp->pgfailback != -FAILBACK_FOLLOWOVER ||
1017             !pp->mpp->pg || !pp->pgindex ||
1018             pp->pgindex != pp->mpp->bestpg)
1019                 return 0;
1020
1021         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1022         vector_foreach_slot(pgp->paths, pp1, i) {
1023                 if (pp1 == pp)
1024                         continue;
1025                 if (pp1->chkrstate != PATH_DOWN && pp1->chkrstate != PATH_SHAKY)
1026                         return 0;
1027         }
1028         return 1;
1029 }
1030
1031 static void
1032 defered_failback_tick (vector mpvec)
1033 {
1034         struct multipath * mpp;
1035         unsigned int i;
1036
1037         vector_foreach_slot (mpvec, mpp, i) {
1038                 /*
1039                  * defered failback getting sooner
1040                  */
1041                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1042                         mpp->failback_tick--;
1043
1044                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1045                                 switch_pathgroup(mpp);
1046                 }
1047         }
1048 }
1049
1050 static void
1051 retry_count_tick(vector mpvec)
1052 {
1053         struct multipath *mpp;
1054         unsigned int i;
1055
1056         vector_foreach_slot (mpvec, mpp, i) {
1057                 if (mpp->retry_tick) {
1058                         mpp->stat_total_queueing_time++;
1059                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1060                         if(--mpp->retry_tick == 0) {
1061                                 dm_queue_if_no_path(mpp->alias, 0);
1062                                 condlog(2, "%s: Disable queueing", mpp->alias);
1063                         }
1064                 }
1065         }
1066 }
1067
1068 int update_prio(struct path *pp, int refresh_all)
1069 {
1070         int oldpriority;
1071         struct path *pp1;
1072         struct pathgroup * pgp;
1073         int i, j, changed = 0;
1074
1075         if (refresh_all) {
1076                 vector_foreach_slot (pp->mpp->pg, pgp, i) {
1077                         vector_foreach_slot (pgp->paths, pp1, j) {
1078                                 oldpriority = pp1->priority;
1079                                 pathinfo(pp1, conf->hwtable, DI_PRIO);
1080                                 if (pp1->priority != oldpriority)
1081                                         changed = 1;
1082                         }
1083                 }
1084                 return changed;
1085         }
1086         oldpriority = pp->priority;
1087         pathinfo(pp, conf->hwtable, DI_PRIO);
1088
1089         if (pp->priority == oldpriority)
1090                 return 0;
1091         return 1;
1092 }
1093
1094 int update_path_groups(struct multipath *mpp, struct vectors *vecs, int refresh)
1095 {
1096         if (reload_map(vecs, mpp, refresh))
1097                 return 1;
1098
1099         dm_lib_release();
1100         if (setup_multipath(vecs, mpp) != 0)
1101                 return 1;
1102         sync_map_state(mpp);
1103
1104         return 0;
1105 }
1106
1107 int
1108 check_path (struct vectors * vecs, struct path * pp)
1109 {
1110         int newstate;
1111         int new_path_up = 0;
1112         int chkr_new_path_up = 0;
1113         int oldchkrstate = pp->chkrstate;
1114
1115         if (!pp->mpp)
1116                 return 0;
1117
1118         if (pp->tick && --pp->tick)
1119                 return 0; /* don't check this path yet */
1120
1121         /*
1122          * provision a next check soonest,
1123          * in case we exit abnormaly from here
1124          */
1125         pp->tick = conf->checkint;
1126
1127         newstate = path_offline(pp);
1128         if (newstate == PATH_REMOVED) {
1129                 condlog(2, "%s: remove path (checker)", pp->dev);
1130                 ev_remove_path(pp, vecs);
1131                 return;
1132         }
1133         if (newstate == PATH_UP)
1134                 newstate = get_state(pp, 1);
1135         else
1136                 checker_clear_message(&pp->checker);
1137
1138         if (newstate == PATH_WILD || newstate == PATH_UNCHECKED) {
1139                 condlog(2, "%s: unusable path", pp->dev);
1140                 pathinfo(pp, conf->hwtable, 0);
1141                 return 1;
1142         }
1143         /*
1144          * Async IO in flight. Keep the previous path state
1145          * and reschedule as soon as possible
1146          */
1147         if (newstate == PATH_PENDING) {
1148                 pp->tick = 1;
1149                 return 0;
1150         }
1151         /*
1152          * Synchronize with kernel state
1153          */
1154         if (update_multipath_strings(pp->mpp, vecs->pathvec)) {
1155                 condlog(1, "%s: Could not synchronize with kernel state",
1156                         pp->dev);
1157                 pp->dmstate = PSTATE_UNDEF;
1158         }
1159         pp->chkrstate = newstate;
1160         if (newstate != pp->state) {
1161                 int oldstate = pp->state;
1162                 pp->state = newstate;
1163
1164                 if (strlen(checker_message(&pp->checker)))
1165                         LOG_MSG(1, checker_message(&pp->checker));
1166
1167                 /*
1168                  * upon state change, reset the checkint
1169                  * to the shortest delay
1170                  */
1171                 pp->checkint = conf->checkint;
1172
1173                 if (newstate == PATH_DOWN || newstate == PATH_SHAKY) {
1174                         /*
1175                          * proactively fail path in the DM
1176                          */
1177                         if (oldstate == PATH_UP ||
1178                             oldstate == PATH_GHOST)
1179                                 fail_path(pp, 1);
1180                         else
1181                                 fail_path(pp, 0);
1182
1183                         /*
1184                          * cancel scheduled failback
1185                          */
1186                         pp->mpp->failback_tick = 0;
1187
1188                         pp->mpp->stat_path_failures++;
1189                         return 1;
1190                 }
1191
1192                 if(newstate == PATH_UP || newstate == PATH_GHOST){
1193                         if ( pp->mpp && pp->mpp->prflag ){
1194                                 /*
1195                                  * Check Persistent Reservation.
1196                                  */
1197                         condlog(2, "%s: checking persistent reservation "
1198                                 "registration", pp->dev);
1199                         mpath_pr_event_handle(pp);
1200                         }
1201                 }
1202
1203                 /*
1204                  * reinstate this path
1205                  */
1206                 if (oldstate != PATH_UP &&
1207                     oldstate != PATH_GHOST)
1208                         reinstate_path(pp, 1);
1209                 else
1210                         reinstate_path(pp, 0);
1211
1212                 new_path_up = 1;
1213
1214                 if (oldchkrstate != PATH_UP && oldchkrstate != PATH_GHOST)
1215                         chkr_new_path_up = 1;
1216
1217                 /*
1218                  * if at least one path is up in a group, and
1219                  * the group is disabled, re-enable it
1220                  */
1221                 if (newstate == PATH_UP)
1222                         enable_group(pp);
1223         }
1224         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1225                 if (pp->dmstate == PSTATE_FAILED ||
1226                     pp->dmstate == PSTATE_UNDEF) {
1227                         /* Clear IO errors */
1228                         reinstate_path(pp, 0);
1229                 } else {
1230                         LOG_MSG(4, checker_message(&pp->checker));
1231                         if (pp->checkint != conf->max_checkint) {
1232                                 /*
1233                                  * double the next check delay.
1234                                  * max at conf->max_checkint
1235                                  */
1236                                 if (pp->checkint < (conf->max_checkint / 2))
1237                                         pp->checkint = 2 * pp->checkint;
1238                                 else
1239                                         pp->checkint = conf->max_checkint;
1240
1241                                 condlog(4, "%s: delay next check %is",
1242                                         pp->dev_t, pp->checkint);
1243                         }
1244                         pp->tick = pp->checkint;
1245                 }
1246         }
1247         else if (newstate == PATH_DOWN &&
1248                  strlen(checker_message(&pp->checker))) {
1249                 if (conf->log_checker_err == LOG_CHKR_ERR_ONCE)
1250                         LOG_MSG(3, checker_message(&pp->checker));
1251                 else
1252                         LOG_MSG(2, checker_message(&pp->checker));
1253         }
1254
1255         pp->state = newstate;
1256
1257         /*
1258          * path prio refreshing
1259          */
1260         condlog(4, "path prio refresh");
1261
1262         if (update_prio(pp, new_path_up) &&
1263             (pp->mpp->pgpolicyfn == (pgpolicyfn *)group_by_prio) &&
1264              pp->mpp->pgfailback == -FAILBACK_IMMEDIATE)
1265                 update_path_groups(pp->mpp, vecs, !new_path_up);
1266         else if (need_switch_pathgroup(pp->mpp, 0)) {
1267                 if (pp->mpp->pgfailback > 0 &&
1268                     (new_path_up || pp->mpp->failback_tick <= 0))
1269                         pp->mpp->failback_tick =
1270                                 pp->mpp->pgfailback + 1;
1271                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE ||
1272                          (chkr_new_path_up && followover_should_failback(pp)))
1273                         switch_pathgroup(pp->mpp);
1274         }
1275         return 1;
1276 }
1277
1278 static void *
1279 checkerloop (void *ap)
1280 {
1281         struct vectors *vecs;
1282         struct path *pp;
1283         int count = 0;
1284         unsigned int i;
1285
1286         mlockall(MCL_CURRENT | MCL_FUTURE);
1287         vecs = (struct vectors *)ap;
1288         condlog(2, "path checkers start up");
1289
1290         /*
1291          * init the path check interval
1292          */
1293         vector_foreach_slot (vecs->pathvec, pp, i) {
1294                 pp->checkint = conf->checkint;
1295         }
1296
1297         while (1) {
1298                 struct timeval diff_time, start_time, end_time;
1299                 int num_paths = 0;
1300
1301                 if (gettimeofday(&start_time, NULL) != 0)
1302                         start_time.tv_sec = 0;
1303                 pthread_cleanup_push(cleanup_lock, &vecs->lock);
1304                 lock(vecs->lock);
1305                 pthread_testcancel();
1306                 condlog(4, "tick");
1307 #ifdef USE_SYSTEMD
1308                 if (conf->watchdog)
1309                         sd_notify(0, "WATCHDOG=1");
1310 #endif
1311                 if (vecs->pathvec) {
1312                         vector_foreach_slot (vecs->pathvec, pp, i) {
1313                                 num_paths += check_path(vecs, pp);
1314                         }
1315                 }
1316                 if (vecs->mpvec) {
1317                         defered_failback_tick(vecs->mpvec);
1318                         retry_count_tick(vecs->mpvec);
1319                 }
1320                 if (count)
1321                         count--;
1322                 else {
1323                         condlog(4, "map garbage collection");
1324                         mpvec_garbage_collector(vecs);
1325                         count = MAPGCINT;
1326                 }
1327
1328                 lock_cleanup_pop(vecs->lock);
1329                 if (start_time.tv_sec &&
1330                     gettimeofday(&end_time, NULL) == 0 &&
1331                     num_paths) {
1332                         timersub(&end_time, &start_time, &diff_time);
1333                         condlog(3, "checked %d path%s in %lu.%06lu secs",
1334                                 num_paths, num_paths > 1 ? "s" : "",
1335                                 diff_time.tv_sec, diff_time.tv_usec);
1336                 }
1337                 sleep(1);
1338         }
1339         return NULL;
1340 }
1341
1342 int
1343 configure (struct vectors * vecs, int start_waiters)
1344 {
1345         struct multipath * mpp;
1346         struct path * pp;
1347         vector mpvec;
1348         int i;
1349
1350         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1351                 return 1;
1352
1353         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1354                 return 1;
1355
1356         if (!(mpvec = vector_alloc()))
1357                 return 1;
1358
1359         /*
1360          * probe for current path (from sysfs) and map (from dm) sets
1361          */
1362         path_discovery(vecs->pathvec, conf, DI_ALL);
1363
1364         vector_foreach_slot (vecs->pathvec, pp, i){
1365                 if (filter_path(conf, pp) > 0){
1366                         vector_del_slot(vecs->pathvec, i);
1367                         free_path(pp);
1368                         i--;
1369                 }
1370                 else
1371                         pp->checkint = conf->checkint;
1372         }
1373         if (map_discovery(vecs))
1374                 return 1;
1375
1376         /*
1377          * create new set of maps & push changed ones into dm
1378          */
1379         if (coalesce_paths(vecs, mpvec, NULL, 1))
1380                 return 1;
1381
1382         /*
1383          * may need to remove some maps which are no longer relevant
1384          * e.g., due to blacklist changes in conf file
1385          */
1386         if (coalesce_maps(vecs, mpvec))
1387                 return 1;
1388
1389         dm_lib_release();
1390
1391         sync_maps_state(mpvec);
1392         vector_foreach_slot(mpvec, mpp, i){
1393                 remember_wwid(mpp->wwid);
1394                 update_map_pr(mpp);
1395         }
1396
1397         /*
1398          * purge dm of old maps
1399          */
1400         remove_maps(vecs);
1401
1402         /*
1403          * save new set of maps formed by considering current path state
1404          */
1405         vector_free(vecs->mpvec);
1406         vecs->mpvec = mpvec;
1407
1408         /*
1409          * start dm event waiter threads for these new maps
1410          */
1411         vector_foreach_slot(vecs->mpvec, mpp, i) {
1412                 if (setup_multipath(vecs, mpp))
1413                         return 1;
1414                 if (start_waiters)
1415                         if (start_waiter_thread(mpp, vecs))
1416                                 return 1;
1417         }
1418         return 0;
1419 }
1420
1421 int
1422 reconfigure (struct vectors * vecs)
1423 {
1424         struct config * old = conf;
1425         int retval = 1;
1426
1427         /*
1428          * free old map and path vectors ... they use old conf state
1429          */
1430         if (VECTOR_SIZE(vecs->mpvec))
1431                 remove_maps_and_stop_waiters(vecs);
1432
1433         if (VECTOR_SIZE(vecs->pathvec))
1434                 free_pathvec(vecs->pathvec, FREE_PATHS);
1435
1436         vecs->pathvec = NULL;
1437         conf = NULL;
1438
1439         /* Re-read any timezone changes */
1440         tzset();
1441
1442         if (!load_config(DEFAULT_CONFIGFILE, udev)) {
1443                 conf->verbosity = old->verbosity;
1444                 conf->daemon = 1;
1445                 configure(vecs, 1);
1446                 free_config(old);
1447                 retval = 0;
1448         }
1449
1450         return retval;
1451 }
1452
1453 static struct vectors *
1454 init_vecs (void)
1455 {
1456         struct vectors * vecs;
1457
1458         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1459
1460         if (!vecs)
1461                 return NULL;
1462
1463         vecs->lock.mutex =
1464                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1465
1466         if (!vecs->lock.mutex)
1467                 goto out;
1468
1469         pthread_mutex_init(vecs->lock.mutex, NULL);
1470         vecs->lock.depth = 0;
1471
1472         return vecs;
1473
1474 out:
1475         FREE(vecs);
1476         condlog(0, "failed to init paths");
1477         return NULL;
1478 }
1479
1480 static void *
1481 signal_set(int signo, void (*func) (int))
1482 {
1483         int r;
1484         struct sigaction sig;
1485         struct sigaction osig;
1486
1487         sig.sa_handler = func;
1488         sigemptyset(&sig.sa_mask);
1489         sig.sa_flags = 0;
1490
1491         r = sigaction(signo, &sig, &osig);
1492
1493         if (r < 0)
1494                 return (SIG_ERR);
1495         else
1496                 return (osig.sa_handler);
1497 }
1498
1499 void
1500 handle_signals(void)
1501 {
1502         if (reconfig_sig && running_state == DAEMON_RUNNING) {
1503                 condlog(2, "reconfigure (signal)");
1504                 pthread_cleanup_push(cleanup_lock,
1505                                 &gvecs->lock);
1506                 lock(gvecs->lock);
1507                 pthread_testcancel();
1508                 reconfigure(gvecs);
1509                 lock_cleanup_pop(gvecs->lock);
1510         }
1511         if (log_reset_sig) {
1512                 condlog(2, "reset log (signal)");
1513                 pthread_mutex_lock(&logq_lock);
1514                 log_reset("multipathd");
1515                 pthread_mutex_unlock(&logq_lock);
1516         }
1517         reconfig_sig = 0;
1518         log_reset_sig = 0;
1519 }
1520
1521 static void
1522 sighup (int sig)
1523 {
1524         reconfig_sig = 1;
1525 }
1526
1527 static void
1528 sigend (int sig)
1529 {
1530         exit_daemon();
1531 }
1532
1533 static void
1534 sigusr1 (int sig)
1535 {
1536         log_reset_sig = 1;
1537 }
1538
1539 static void
1540 sigusr2 (int sig)
1541 {
1542         condlog(3, "SIGUSR2 received");
1543 }
1544
1545 static void
1546 signal_init(void)
1547 {
1548         sigset_t set;
1549
1550         sigemptyset(&set);
1551         sigaddset(&set, SIGHUP);
1552         sigaddset(&set, SIGUSR1);
1553         sigaddset(&set, SIGUSR2);
1554         pthread_sigmask(SIG_BLOCK, &set, NULL);
1555
1556         signal_set(SIGHUP, sighup);
1557         signal_set(SIGUSR1, sigusr1);
1558         signal_set(SIGUSR2, sigusr2);
1559         signal_set(SIGINT, sigend);
1560         signal_set(SIGTERM, sigend);
1561         signal(SIGPIPE, SIG_IGN);
1562 }
1563
1564 static void
1565 setscheduler (void)
1566 {
1567         int res;
1568         static struct sched_param sched_param = {
1569                 .sched_priority = 99
1570         };
1571
1572         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1573
1574         if (res == -1)
1575                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1576         return;
1577 }
1578
1579 static void
1580 set_oom_adj (void)
1581 {
1582 #ifdef OOM_SCORE_ADJ_MIN
1583         int retry = 1;
1584         char *file = "/proc/self/oom_score_adj";
1585         int score = OOM_SCORE_ADJ_MIN;
1586 #else
1587         int retry = 0;
1588         char *file = "/proc/self/oom_adj";
1589         int score = OOM_ADJUST_MIN;
1590 #endif
1591         FILE *fp;
1592         struct stat st;
1593         char *envp;
1594
1595         envp = getenv("OOMScoreAdjust");
1596         if (envp) {
1597                 condlog(3, "Using systemd provided OOMScoreAdjust");
1598                 return;
1599         }
1600         do {
1601                 if (stat(file, &st) == 0){
1602                         fp = fopen(file, "w");
1603                         if (!fp) {
1604                                 condlog(0, "couldn't fopen %s : %s", file,
1605                                         strerror(errno));
1606                                 return;
1607                         }
1608                         fprintf(fp, "%i", score);
1609                         fclose(fp);
1610                         return;
1611                 }
1612                 if (errno != ENOENT) {
1613                         condlog(0, "couldn't stat %s : %s", file,
1614                                 strerror(errno));
1615                         return;
1616                 }
1617 #ifdef OOM_ADJUST_MIN
1618                 file = "/proc/self/oom_adj";
1619                 score = OOM_ADJUST_MIN;
1620 #else
1621                 retry = 0;
1622 #endif
1623         } while (retry--);
1624         condlog(0, "couldn't adjust oom score");
1625 }
1626
1627 static int
1628 child (void * param)
1629 {
1630         pthread_t check_thr, uevent_thr, uxlsnr_thr, uevq_thr;
1631         pthread_attr_t log_attr, misc_attr, uevent_attr;
1632         struct vectors * vecs;
1633         struct multipath * mpp;
1634         int i;
1635 #ifdef USE_SYSTEMD
1636         unsigned long checkint;
1637 #endif
1638         int rc, pid_rc;
1639         char *envp;
1640
1641         mlockall(MCL_CURRENT | MCL_FUTURE);
1642         sem_init(&exit_sem, 0, 0);
1643         signal_init();
1644
1645         udev = udev_new();
1646
1647         setup_thread_attr(&misc_attr, 64 * 1024, 1);
1648         setup_thread_attr(&uevent_attr, 128 * 1024, 1);
1649         setup_thread_attr(&waiter_attr, 32 * 1024, 1);
1650
1651         if (logsink == 1) {
1652                 setup_thread_attr(&log_attr, 64 * 1024, 0);
1653                 log_thread_start(&log_attr);
1654                 pthread_attr_destroy(&log_attr);
1655         }
1656
1657         running_state = DAEMON_START;
1658
1659 #ifdef USE_SYSTEMD
1660         sd_notify(0, "STATUS=startup");
1661 #endif
1662         condlog(2, "--------start up--------");
1663         condlog(2, "read " DEFAULT_CONFIGFILE);
1664
1665         if (load_config(DEFAULT_CONFIGFILE, udev))
1666                 goto failed;
1667
1668         if (init_checkers()) {
1669                 condlog(0, "failed to initialize checkers");
1670                 goto failed;
1671         }
1672         if (init_prio()) {
1673                 condlog(0, "failed to initialize prioritizers");
1674                 goto failed;
1675         }
1676
1677         setlogmask(LOG_UPTO(conf->verbosity + 3));
1678
1679         envp = getenv("LimitNOFILE");
1680
1681         if (envp) {
1682                 condlog(2,"Using systemd provided open fds limit of %s", envp);
1683         } else if (conf->max_fds) {
1684                 struct rlimit fd_limit;
1685
1686                 if (getrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1687                         condlog(0, "can't get open fds limit: %s",
1688                                 strerror(errno));
1689                         fd_limit.rlim_cur = 0;
1690                         fd_limit.rlim_max = 0;
1691                 }
1692                 if (fd_limit.rlim_cur < conf->max_fds) {
1693                         fd_limit.rlim_cur = conf->max_fds;
1694                         if (fd_limit.rlim_max < conf->max_fds)
1695                                 fd_limit.rlim_max = conf->max_fds;
1696                         if (setrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1697                                 condlog(0, "can't set open fds limit to "
1698                                         "%lu/%lu : %s",
1699                                         fd_limit.rlim_cur, fd_limit.rlim_max,
1700                                         strerror(errno));
1701                         } else {
1702                                 condlog(3, "set open fds limit to %lu/%lu",
1703                                         fd_limit.rlim_cur, fd_limit.rlim_max);
1704                         }
1705                 }
1706
1707         }
1708
1709         vecs = gvecs = init_vecs();
1710         if (!vecs)
1711                 goto failed;
1712
1713         setscheduler();
1714         set_oom_adj();
1715
1716         conf->daemon = 1;
1717         udev_set_sync_support(0);
1718 #ifdef USE_SYSTEMD
1719         envp = getenv("WATCHDOG_USEC");
1720         if (envp && sscanf(envp, "%lu", &checkint) == 1) {
1721                 /* Value is in microseconds */
1722                 conf->max_checkint = checkint / 1000000;
1723                 /* Rescale checkint */
1724                 if (conf->checkint > conf->max_checkint)
1725                         conf->checkint = conf->max_checkint;
1726                 else
1727                         conf->checkint = conf->max_checkint / 4;
1728                 condlog(3, "enabling watchdog, interval %d max %d",
1729                         conf->checkint, conf->max_checkint);
1730                 conf->watchdog = conf->checkint;
1731         }
1732 #endif
1733         /*
1734          * Start uevent listener early to catch events
1735          */
1736         if ((rc = pthread_create(&uevent_thr, &uevent_attr, ueventloop, udev))) {
1737                 condlog(0, "failed to create uevent thread: %d", rc);
1738                 goto failed;
1739         }
1740         pthread_attr_destroy(&uevent_attr);
1741         if ((rc = pthread_create(&uxlsnr_thr, &misc_attr, uxlsnrloop, vecs))) {
1742                 condlog(0, "failed to create cli listener: %d", rc);
1743                 goto failed;
1744         }
1745         /*
1746          * fetch and configure both paths and multipaths
1747          */
1748 #ifdef USE_SYSTEMD
1749         sd_notify(0, "STATUS=configure");
1750 #endif
1751         running_state = DAEMON_CONFIGURE;
1752
1753         lock(vecs->lock);
1754         if (configure(vecs, 1)) {
1755                 unlock(vecs->lock);
1756                 condlog(0, "failure during configuration");
1757                 goto failed;
1758         }
1759         unlock(vecs->lock);
1760
1761         /*
1762          * start threads
1763          */
1764         if ((rc = pthread_create(&check_thr, &misc_attr, checkerloop, vecs))) {
1765                 condlog(0,"failed to create checker loop thread: %d", rc);
1766                 goto failed;
1767         }
1768         if ((rc = pthread_create(&uevq_thr, &misc_attr, uevqloop, vecs))) {
1769                 condlog(0, "failed to create uevent dispatcher: %d", rc);
1770                 goto failed;
1771         }
1772         pthread_attr_destroy(&misc_attr);
1773
1774         /* Startup complete, create logfile */
1775         pid_rc = pidfile_create(DEFAULT_PIDFILE, daemon_pid);
1776         /* Ignore errors, we can live without */
1777
1778         running_state = DAEMON_RUNNING;
1779 #ifdef USE_SYSTEMD
1780         sd_notify(0, "READY=1\nSTATUS=running");
1781 #endif
1782
1783         /*
1784          * exit path
1785          */
1786         while(sem_wait(&exit_sem) != 0); /* Do nothing */
1787
1788 #ifdef USE_SYSTEMD
1789         sd_notify(0, "STATUS=shutdown");
1790 #endif
1791         running_state = DAEMON_SHUTDOWN;
1792         lock(vecs->lock);
1793         if (conf->queue_without_daemon == QUE_NO_DAEMON_OFF)
1794                 vector_foreach_slot(vecs->mpvec, mpp, i)
1795                         dm_queue_if_no_path(mpp->alias, 0);
1796         remove_maps_and_stop_waiters(vecs);
1797         unlock(vecs->lock);
1798
1799         pthread_cancel(check_thr);
1800         pthread_cancel(uevent_thr);
1801         pthread_cancel(uxlsnr_thr);
1802         pthread_cancel(uevq_thr);
1803
1804         lock(vecs->lock);
1805         free_pathvec(vecs->pathvec, FREE_PATHS);
1806         vecs->pathvec = NULL;
1807         unlock(vecs->lock);
1808         /* Now all the waitevent threads will start rushing in. */
1809         while (vecs->lock.depth > 0) {
1810                 sleep (1); /* This is weak. */
1811                 condlog(3, "Have %d wait event checkers threads to de-alloc,"
1812                         " waiting...", vecs->lock.depth);
1813         }
1814         pthread_mutex_destroy(vecs->lock.mutex);
1815         FREE(vecs->lock.mutex);
1816         vecs->lock.depth = 0;
1817         vecs->lock.mutex = NULL;
1818         FREE(vecs);
1819         vecs = NULL;
1820
1821         cleanup_checkers();
1822         cleanup_prio();
1823
1824         dm_lib_release();
1825         dm_lib_exit();
1826
1827         /* We're done here */
1828         if (!pid_rc) {
1829                 condlog(3, "unlink pidfile");
1830                 unlink(DEFAULT_PIDFILE);
1831         }
1832
1833         condlog(2, "--------shut down-------");
1834
1835         if (logsink == 1)
1836                 log_thread_stop();
1837
1838         /*
1839          * Freeing config must be done after condlog() and dm_lib_exit(),
1840          * because logging functions like dlog() and dm_write_log()
1841          * reference the config.
1842          */
1843         free_config(conf);
1844         conf = NULL;
1845         udev_unref(udev);
1846         udev = NULL;
1847 #ifdef _DEBUG_
1848         dbg_free_final(NULL);
1849 #endif
1850
1851 #ifdef USE_SYSTEMD
1852         sd_notify(0, "ERRNO=0");
1853 #endif
1854         exit(0);
1855
1856 failed:
1857 #ifdef USE_SYSTEMD
1858         sd_notify(0, "ERRNO=1");
1859 #endif
1860         exit(1);
1861 }
1862
1863 static int
1864 daemonize(void)
1865 {
1866         int pid;
1867         int dev_null_fd;
1868
1869         if( (pid = fork()) < 0){
1870                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1871                 return -1;
1872         }
1873         else if (pid != 0)
1874                 return pid;
1875
1876         setsid();
1877
1878         if ( (pid = fork()) < 0)
1879                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1880         else if (pid != 0)
1881                 _exit(0);
1882
1883         if (chdir("/") < 0)
1884                 fprintf(stderr, "cannot chdir to '/', continuing\n");
1885
1886         dev_null_fd = open("/dev/null", O_RDWR);
1887         if (dev_null_fd < 0){
1888                 fprintf(stderr, "cannot open /dev/null for input & output : %s\n",
1889                         strerror(errno));
1890                 _exit(0);
1891         }
1892
1893         close(STDIN_FILENO);
1894         if (dup(dev_null_fd) < 0) {
1895                 fprintf(stderr, "cannot dup /dev/null to stdin : %s\n",
1896                         strerror(errno));
1897                 _exit(0);
1898         }
1899         close(STDOUT_FILENO);
1900         if (dup(dev_null_fd) < 0) {
1901                 fprintf(stderr, "cannot dup /dev/null to stdout : %s\n",
1902                         strerror(errno));
1903                 _exit(0);
1904         }
1905         close(STDERR_FILENO);
1906         if (dup(dev_null_fd) < 0) {
1907                 fprintf(stderr, "cannot dup /dev/null to stderr : %s\n",
1908                         strerror(errno));
1909                 _exit(0);
1910         }
1911         close(dev_null_fd);
1912         daemon_pid = getpid();
1913         return 0;
1914 }
1915
1916 int
1917 main (int argc, char *argv[])
1918 {
1919         extern char *optarg;
1920         extern int optind;
1921         int arg;
1922         int err;
1923
1924         logsink = 1;
1925         running_state = DAEMON_INIT;
1926         dm_init();
1927
1928         if (getuid() != 0) {
1929                 fprintf(stderr, "need to be root\n");
1930                 exit(1);
1931         }
1932
1933         /* make sure we don't lock any path */
1934         if (chdir("/") < 0)
1935                 fprintf(stderr, "can't chdir to root directory : %s\n",
1936                         strerror(errno));
1937         umask(umask(077) | 022);
1938
1939         conf = alloc_config();
1940
1941         if (!conf)
1942                 exit(1);
1943
1944         while ((arg = getopt(argc, argv, ":dsv:k::")) != EOF ) {
1945         switch(arg) {
1946                 case 'd':
1947                         logsink = 0;
1948                         //debug=1; /* ### comment me out ### */
1949                         break;
1950                 case 'v':
1951                         if (sizeof(optarg) > sizeof(char *) ||
1952                             !isdigit(optarg[0]))
1953                                 exit(1);
1954
1955                         conf->verbosity = atoi(optarg);
1956                         break;
1957                 case 's':
1958                         logsink = -1;
1959                         break;
1960                 case 'k':
1961                         uxclnt(optarg);
1962                         exit(0);
1963                 default:
1964                         ;
1965                 }
1966         }
1967         if (optind < argc) {
1968                 char cmd[CMDSIZE];
1969                 char * s = cmd;
1970                 char * c = s;
1971
1972                 while (optind < argc) {
1973                         if (strchr(argv[optind], ' '))
1974                                 c += snprintf(c, s + CMDSIZE - c, "\"%s\" ", argv[optind]);
1975                         else
1976                                 c += snprintf(c, s + CMDSIZE - c, "%s ", argv[optind]);
1977                         optind++;
1978                 }
1979                 c += snprintf(c, s + CMDSIZE - c, "\n");
1980                 uxclnt(s);
1981                 exit(0);
1982         }
1983
1984         if (logsink < 1)
1985                 err = 0;
1986         else
1987                 err = daemonize();
1988
1989         if (err < 0)
1990                 /* error */
1991                 exit(1);
1992         else if (err > 0)
1993                 /* parent dies */
1994                 exit(0);
1995         else
1996                 /* child lives */
1997                 return (child(NULL));
1998 }
1999
2000 void *  mpath_pr_event_handler_fn (void * pathp )
2001 {
2002         struct multipath * mpp;
2003         int i,j, ret, isFound;
2004         struct path * pp = (struct path *)pathp;
2005         unsigned char *keyp;
2006         uint64_t prkey;
2007         struct prout_param_descriptor *param;
2008         struct prin_resp *resp;
2009
2010         mpp = pp->mpp;
2011
2012         resp = mpath_alloc_prin_response(MPATH_PRIN_RKEY_SA);
2013         if (!resp){
2014                 condlog(0,"%s Alloc failed for prin response", pp->dev);
2015                 return NULL;
2016         }
2017
2018         ret = prin_do_scsi_ioctl(pp->dev, MPATH_PRIN_RKEY_SA, resp, 0);
2019         if (ret != MPATH_PR_SUCCESS )
2020         {
2021                 condlog(0,"%s : pr in read keys service action failed. Error=%d", pp->dev, ret);
2022                 goto out;
2023         }
2024
2025         condlog(3, " event pr=%d addlen=%d",resp->prin_descriptor.prin_readkeys.prgeneration,
2026                         resp->prin_descriptor.prin_readkeys.additional_length );
2027
2028         if (resp->prin_descriptor.prin_readkeys.additional_length == 0 )
2029         {
2030                 condlog(1, "%s: No key found. Device may not be registered.", pp->dev);
2031                 ret = MPATH_PR_SUCCESS;
2032                 goto out;
2033         }
2034         prkey = 0;
2035         keyp = (unsigned char *)mpp->reservation_key;
2036         for (j = 0; j < 8; ++j) {
2037                 if (j > 0)
2038                         prkey <<= 8;
2039                 prkey |= *keyp;
2040                 ++keyp;
2041         }
2042         condlog(2, "Multipath  reservation_key: 0x%" PRIx64 " ", prkey);
2043
2044         isFound =0;
2045         for (i = 0; i < resp->prin_descriptor.prin_readkeys.additional_length/8; i++ )
2046         {
2047                 condlog(2, "PR IN READKEYS[%d]  reservation key:",i);
2048                 dumpHex((char *)&resp->prin_descriptor.prin_readkeys.key_list[i*8], 8 , -1);
2049                 if (!memcmp(mpp->reservation_key, &resp->prin_descriptor.prin_readkeys.key_list[i*8], 8))
2050                 {
2051                         condlog(2, "%s: pr key found in prin readkeys response", mpp->alias);
2052                         isFound =1;
2053                         break;
2054                 }
2055         }
2056         if (!isFound)
2057         {
2058                 condlog(0, "%s: Either device not registered or ", pp->dev);
2059                 condlog(0, "host is not authorised for registration. Skip path");
2060                 ret = MPATH_PR_OTHER;
2061                 goto out;
2062         }
2063
2064         param= malloc(sizeof(struct prout_param_descriptor));
2065         memset(param, 0 , sizeof(struct prout_param_descriptor));
2066
2067         for (j = 7; j >= 0; --j) {
2068                 param->sa_key[j] = (prkey & 0xff);
2069                 prkey >>= 8;
2070         }
2071         param->num_transportid = 0;
2072
2073         condlog(3, "device %s:%s", pp->dev, pp->mpp->wwid);
2074
2075         ret = prout_do_scsi_ioctl(pp->dev, MPATH_PROUT_REG_IGN_SA, 0, 0, param, 0);
2076         if (ret != MPATH_PR_SUCCESS )
2077         {
2078                 condlog(0,"%s: Reservation registration failed. Error: %d", pp->dev, ret);
2079         }
2080         mpp->prflag = 1;
2081
2082         free(param);
2083 out:
2084         free(resp);
2085         return NULL;
2086 }
2087
2088 int mpath_pr_event_handle(struct path *pp)
2089 {
2090         pthread_t thread;
2091         int rc;
2092         pthread_attr_t attr;
2093         struct multipath * mpp;
2094
2095         mpp = pp->mpp;
2096
2097         if (!mpp->reservation_key)
2098                 return -1;
2099
2100         pthread_attr_init(&attr);
2101         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
2102
2103         rc = pthread_create(&thread, NULL , mpath_pr_event_handler_fn, pp);
2104         if (rc) {
2105                 condlog(0, "%s: ERROR; return code from pthread_create() is %d", pp->dev, rc);
2106                 return -1;
2107         }
2108         pthread_attr_destroy(&attr);
2109         rc = pthread_join(thread, NULL);
2110         return 0;
2111 }
2112